Chemical principles dictate that the specific binding of a target to its complementary probes on a DNA microarray surface, and the nonspecific binding between other nucleotide segments and the same probes, are mutually competitive. We demonstrate that this mechanism can be understood by considering the competitive chemical reaction taking place on the microarray surface. Inspired by the pioneering work of Zhang and Hekstra, we have developed a physical model for microarray signal analysis, based on possible reaction mechanisms, and implemented it with a parallel, generic, simulated-annealing algorithm. Using data supplied by the Affymetrix Latin-square spike-in experiments, our model showed excellent fitting of the data. This correlation between the predicted expression levels and the spike-in concentrations of test transcripts demonstrated good predictive abilities of our model.